Probabilistic Extension of Neuro-Symbolic AGI Robots based on Belnap's Typed Intensional FOL

Zoran Majkic

REVIEW 3 major objections 4 minor 47 references

Probabilistic Extension of Neuro-Symbolic AGI Robots based on Belnap's Typed Intensional FOL

T0 review · 3 major / 4 minor · reviewed 2026-08-02 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read This paper claims that a four-valued intensional logic for autonomous robots can be extended with a probability layer that assigns probabilities to currently unknown sentences without changing the knowledge base or its logical consequences.

desk verdict A speculative adaptation of Nilsson's semantics to a 4-valued intensional FOL, but the preservation results are tautologies and the finite-sample-space assumption doesn't survive the infinite Herbrand base. read the letter →

arxiv 2607.13073 v2 pith:J5SRVRWP submitted 2026-07-12 cs.AI

Zoran Majkic This is my paper

classification cs.AI MSC 03B4803B5068T27

keywords probabilisticlogicneuro-symbolicAGIintensionalfirst-orderfour-valuedpossible-worldssemanticsmaximumentropyclosedknowledgeassumptionautoepistemicreasoning

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper tries to give IFOLB, a four-valued intensional first-order logic for neuro-symbolic robots, a probabilistic layer. The logic already distinguishes true, false, unknown, and inconsistent; the goal is to let a robot compute probabilities for sentences it currently marks unknown, so decisions under uncertainty can be made inside the same logical framework. The construction builds a probability space whose sample points are the possible worlds—assignments of the four truth values to all ground atomic facts—that agree with the current knowledge base on every atom whose value is known. For any unknown sentence, its probability of taking a given truth value is the total probability mass of those worlds in which it takes that value; the distribution is chosen by maximum information entropy. A local version restricts the same construction to the few predicates involved in a concrete decision problem, which the paper argues makes real-time use feasible.

What carries the argument

The load-bearing object is the global sample space S: the set of possible worlds (assignments of the four truth values to all ground atomic facts) that agree with the current knowledge base on every atom whose value is known. The probability of an unknown sentence φ having truth value u is computed by summing the density KI over the worlds in S where φ takes value u. The local machinery replaces S with S_φ, the analogous set built only from the ground atoms of the predicates in φ, and the preservation argument is the same: constrained atoms get probability 1. The entropy objective H(KI) = −Σ KI(v) log KI(v) is what makes the distribution the least committal one.

What would settle it

Take a language with one constant, one unary function symbol, and one predicate, so the base of ground atoms is infinite and the global sample space S is the set of all four-valued assignments to this infinite set. With the knowledge base empty, the maximum-entropy density KI and its entropy sum are not finitely defined; a direct calculation of p_u for a sentence such as ∀x.P(x) then has no well-defined value, contradicting the claim that the method computes probabilities for unknown sentences. Confirming this divergence would settle the matter.

Watch

Extended reading notes

Core claim

On its own terms, the paper's central claim is that the set S of possible worlds consistent with the current knowledge base K defines a probability space that preserves K and all deductions from it. For any sentence φ with current truth value unknown, the paper defines p_u = Σ_{v∈S, v*(φ)=u} KI(v) as the probability that φ takes truth value u ∈ {true, false, unknown, inconsistent}; the first corollary shows known atoms get probability 1, so the database is invariant, and the wN meta-predicate records these probabilistic facts as ordinary first-order atoms. The same construction is repeated locally over S_φ, the sample space built only from the ground atoms of the predicates occurring in φ, w

Load-bearing premise

The paper assumes that the truth value of a sentence—especially a quantified sentence—is determined by the ground atoms in a finite local (or even global) sample space, so that probabilities computed over S or S_φ match the logic's full semantics; this compatibility is asserted rather than proved.

Editorial extensions

If this is right

A robot can attach probabilities to unknown facts, so planning under uncertainty can stay inside the four-valued logic instead of dropping to a two-valued approximation.
Because known atoms keep probability 1, the knowledge base and its deductive closure are preserved when probabilistic knowledge is added.
The local structure lets a robot compute probabilities for a subproblem using only the predicates in that subproblem, which is what the paper says makes real-time decisions feasible.
Probabilistic facts are expressed as ordinary atoms of the wN meta-predicate, so they can be stored, introspected via the Know predicate, and used in the robot's knowledge-based reasoning and explanations.
The maximum-entropy choice guarantees that the probabilities introduce no structure beyond what the robot already knows, making the assignment the most prudent one.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

The paper's finiteness assumption is doing more work than stated: for a first-order language with one constant and one unary function symbol, the base of ground atoms is infinite and the global sample space S is uncountable, so the finite probability space and the entropy sum are not defined without an additional limiting argument.
If the local construction is meant to approximate the global one, a direct check on a quantified sentence whose truth depends on ground atoms outside the local sample space would show whether the local probability formula can disagree with the global one; the paper leaves this compatibility unproved.
The symmetry-group framing suggests a modular implementation: each local decision adds a new element to the group and the combined probabilistic knowledge is a disjoint union of independently computed local distributions. Such an implementation would be testable in a small robotics domain with a handful of predicates.
The paper is agnostic about how the neural network optimizes the entropy, so any differentiable optimizer that respects the constraints—for example, a softmax layer with a constraint-aware loss—can serve as the probability engine; that makes the framework implementable without altering the logical layer.

Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Desk Editor's Note

read the letter

Two things to know. This is a speculative theory paper that wraps Nilsson's possible-world probability semantics in the author's 4-valued intensional logic. The central preservation results, Corollaries 1 and 2, are true by definition, not theorems, and the paper never reconciles its finite-sample-space assumption with the potentially infinite Herbrand base. So the probabilistic extension as stated does not land for full IFOLB.

What's genuinely useful: the paper identifies a real gap — how to attach probabilities to unknown sentences in a logic that already has truth values for unknown and inconsistent — and it proposes a principled way: take possible worlds as Herbrand interpretations that agree with the current knowledge base on all known atoms, then sum distributions over those worlds. The local restriction to a small predicate subset for real-time decisions is a sensible factorization idea, and the author correctly cites Nilsson and acknowledges that the framework is theoretical. The intention is honest.

The soft spots are load-bearing. First, in (20) the sample space S is defined as {v : v(A)=v(A) whenever v(A)≠⊥}, so the probability constraints (19) are satisfied automatically for any distribution over S. Preservation of the knowledge base is a definitional choice, not a derived invariant. Second, Definition 7 does not guarantee S is finite. Section 2 says the paper only treats finite sample spaces, but the Herbrand base for IFOLB includes abstracted ground terms and is generally countably infinite, making the sums in (22) and the entropy in (24) undefined. The local construction restricts to finite Hφ, but that does not determine the truth value of quantified sentences in the original semantics. Equation (27) therefore assigns a probability to an object that isn't a function of the restricted world. The reader's stress-test note is on target. Third, the neural-network computation of the max-entropy distribution is asserted without architecture, data, or experimental analysis. Finally, the paper leans heavily on an extensive series of the author's own prior papers for the underlying IFOLB, which makes independent checking laborious, and the writing is repetitive and marketing-toned in the comparisons to LLMs and DeepProbLog.

Who gets value: someone working on many-valued probabilistic semantics might read this as a cautionary example, and the author's earlier IFOLB work is clearly substantive. But this paper does not add a sound formal layer on top of it.

Recommendation: I would not desk-reject it; the topic is relevant and the definitions are concrete enough that a knowledgeable referee can give specific feedback. But I'd expect major revision, and I wouldn't bet on the central construction surviving contact with the finiteness and quantification issues. It's a maybe for a reading group, and I wouldn't cite it as a reference for probabilistic IFOLB in its current form.

Referee Report

3 major / 4 minor

Summary. The paper proposes a probabilistic extension of the four-valued Belnap-style intensional first-order logic IFOL_B. The central construction is a Nilsson-style probability structure over Herbrand interpretations: the global sample space S in Definition 7 consists of all Herbrand interpretations that agree with the current interpretation v on all atoms whose truth value is already known (not ⊥), and the local sample space S_φ in Definition 8 restricts to a finite Herbrand base H_φ for a subproblem sentence φ. The paper defines a probability density KI over S (or S_φ), imposes constraints that known atoms keep probability 1, and introduces a meta-predicate wN so that wN(⋖φ⋗, u, p) asserts that the probability that φ has truth value u is p. It claims that this construction is a global/local symmetry transformation preserving the current knowledge database K and logical deduction, that Corollaries 1 and 2 establish consistency with K, and that neural networks can compute the maximum-entropy KI. The intended contribution is to give IFOL_B a probabilistic layer for unknown and inconsistent sentences, with application to robot decision-making.

Significance. If the construction were mathematically sound, the paper would address a real gap in neuro-symbolic systems: combining a four-valued autoepistemic logic with Nilsson-style possible-world probabilities, while distinguishing global from local inference. The idea of reifying probabilistic assertions through an intensional predicate wN is coherent, and the use of maximum entropy is a reasonable design choice for a least-committal distribution. However, the current manuscript does not deliver a well-defined formal system. The sample space is allowed to be infinite while all probability equations are only defined for finite spaces; the local restriction does not determine truth values of quantified sentences; and the preservation theorems are consequences of the definitions rather than substantive results. The paper also contains no machine-checked proofs, no implementation, and no empirical validation, so the practical claims about neural computation of KI remain unsupported. Because the central formal claims are not established, the significance of the contribution as stated cannot be assessed.

major comments (3)

[§2, Definition 7, Definition 8, Eqs. (20), (22), (24)] The probability formalism is only defined for finite sample spaces, but the paper never shows that the global sample space S in Eq. (20) is finite. Section 2 explicitly restricts to finite S ("In what follows we will consider only finite sample space S"), yet Definition 7 defines S as a subset of IH ⊂ X^H, where H is the Herbrand base of IFOL_B. H contains iterated abstracted terms such as ⋖P(a)⋗ and ⋖P(⋖P(a)⋗)⋗, so H is generally countably infinite and X^H is uncountable; S is typically infinite too. In that case the sum in Eq. (22) over {v ∈ S : v*(φ)=u} is not a well-defined finite sum, and the entropy in Eq. (24) is not defined on an infinite set without additional measure-theoretic structure. The local construction in Definition 8 does not fix this: even if H_φ is finite, a function v : H_φ → X does not in general determine v*(ψ) for a quantified sentence ψ, because quantifiers rang
[Definition 6, Definition 7, Corollaries 1 and 2, Eq. (19)] Corollaries 1 and 2 are tautological rather than derived invariance results. In Definition 7, the sample space S in Eq. (20) is defined as the set of interpretations v that agree with the current v on every atom A with v(A) ≠ ⊥. Therefore, for any such known atom A with u = v(A) ≠ ⊥, the set {v ∈ S : v(A) = u} is exactly S, and Eq. (19) reduces to Σ_{v∈S} KI(v) = 1, which is just normalization. The same holds for the local S_φ in Eq. (25) and Eq. (26). The paper presents this as "preserving current knowledge database K and logical deduction," but preservation of K is built into the definition of the sample space, not a consequence of any probabilistic or logical principle. Moreover, preservation of K alone does not imply preservation of the deductive closure under IFOL_B's autoepistemic axioms; the paper gives no argument that adding probabilistic facts wN leaves all deductions invariant
[Eq. (20) notation and Definition 8] Equation (20) uses the same symbol v for the current Herbrand interpretation and for the arbitrary element of S: S =def {v ∈ IH ⊂ X^H | for every A ∈ H, v(A) = v(A) if v(A) ≠ ⊥}. As written, this is a tautological condition unless the two occurrences are intended to be different symbols. If they are intended to be different, the paper must use distinct names (e.g., v and v') and clarify that the condition quantifies over all A ∈ H. The same issue appears in Eq. (25). This may seem cosmetic, but it obscures the actual content of the definition and contributes to the impression that Corollaries 1 and 2 are vacuous.

minor comments (4)

[Section 2, Eq. (24)] The entropy expression H(KI) = -Σ_{v∈S} KI(v)·log KI(v) does not specify the base of the logarithm, and no convention is given for the term KI(v)=0. These are minor but should be stated for a formal definition.
[Corollary 2 proof] The proof says "The local symmetry transformation Sφ has no effects, from (25), for the ground atoms in H which are not in Hφ." But Sφ is only defined on Hφ, so it does not assign truth values to atoms outside Hφ. The statement is informal and should be rephrased in terms of the extension used in Eq. (27).
[Appendix B, Section 7] The heading "Appendix B: Neuro-symbolic Archetuture" is followed by no text. Either add the content or remove the empty section.
[Throughout] There are numerous typos and infelicities, e.g., "Sistem 1" for "System 1" and "Archetuture" for "Architecture." The paper would benefit from careful proofreading.

Circularity Check

2 steps flagged · score 6.0 of 10

The claimed global symmetry that 'preserves K' is definitional: Eq. (20) is the preservation constraint rewritten as the sample space, so Corollary 1 is a tautology rather than a derived invariant.

self definitional [Definition 7, Eq. (20); Corollary 1]
"S =def {v ∈ IH ⊂XH| forevery A ∈H,v (A) = v(A) if v(A) ̸= ⊥} ... pu = ∑_{v∈S,v(A)=u}KI(v) = ∑_{v∈S}KI(v) from (20) =µ(S) = 1. which preserves robot's current knowledge database and all logical deductions from them."

Eq. (19) fixes the probability of every known atom to 1; Eq. (20) defines S as the set of worlds already agreeing with the current v on exactly those atoms. Corollary 1's proof substitutes (20) into the sum and obtains 1. Thus the 'global symmetry' that preserves K is not an invariant derived from a transformation; it is the defining condition of the sample space. The same pattern is repeated locally in Definition 8/Corollary 2.
self definitional [Definition 8, Eq. (25); Corollary 2]
"Sφ =def {v :Hφ →X| forevery A ∈Hφ ⊂H,v (A) = v(A) if v(A) ̸= ⊥} ... pu = ∑_{v∈Sφ,v(A)=u}KIφ(v) = ∑_{v∈Sφ}KIφ(v) from (25) =µφ(S) = 1. which preserves robot's current knowledge database and all logical deductions from them."

The local sample space is restricted to functions that already assign the database values to every known atom in Hφ; the local probability-density constraint (26) is then satisfied by construction. The 'proof' in Corollary 2 is a restatement of the defining property of Sφ, not an independent preservation result.

full rationale

The paper has an independent probabilistic core: one chooses a max-entropy KI over a sample space and computes p_u as the KI-mass of worlds making an unknown sentence true. That part is not a renamed fit of the target sentence. However, the abstract's advertised 'global symmetry transformation that preserves K' is circular in a precise, quotable sense: Eq. (20) defines S as the set of Herbrand interpretations that already agree with K on every known atom, which is exactly the preservation constraint (19). Corollary 1 then 'proves' preservation by substituting the definition; Corollary 2 repeats the same move locally. Because this definitional conservation is one of the paper's central claims, the score is 6 rather than lower. The author's heavy reliance on prior self-citations [2,6,32] for IFOLB, Know, and autoepistemic axioms is background self-citation but is not itself scored as circularity here, and the separate finite-sample-space/quantifier mismatch is a correctness problem, not a circularity, so it is not used to raise the score.

Assumptions & free parameters 0 free parameters · 5 assumptions · 1 invented entities

The paper introduces no fitted numerical parameters. It depends on the author's prior IFOL_B framework for foundational consistency, on standard Herbrand model theory, on a finite-sample-space assumption, on the existence of a neural max-entropy solver, and on the semantic compatibility of local predicate restrictions with quantification.

assumptions (5)

domain assumption IFOL_B with Belnap's bilattice and autoepistemic axioms is consistent and supports the deductive machinery used here.
Section 1 and throughout the paper cite [6,32] for the underlying logic; the paper does not re-prove these foundations.
standard math The Herbrand interpretation v extends uniquely to all sentences v*: L0 -> X and the set of well-defined interpretations is nonempty.
Used in (1)-(2) and in the definition of sample spaces; standard for Herbrand interpretations, but assumes built-in predicates are respected.
domain assumption The sample space S (and S_φ) is finite, so the finite-additivity form of Kolmogorov's axioms applies.
Section 2 states 'we will consider only finite sample space S', but Definition 7 uses the Herbrand base H, which may be infinite; never reconciled.
ad hoc to paper A maximum-entropy probability distribution KI exists and can be computed by a neural network subject to constraints (19).
Section 3 asserts this without providing an architecture, training procedure, or proof of convergence/constraint satisfaction.
ad hoc to paper Restricting the Herbrand interpretation to H_φ and extending it to sentences preserves the truth values (including quantifiers) used in the local probability computation.
Definition 8 and equation (27) assume that atoms outside H_φ do not affect the probability of ψ(x)/g; this is not proved and fails for sentences like ∃xτ.pi(...) that range over the full domain.

invented entities (1)

wN (ternary probabilistic meta-predicate)
purpose: Represents facts of the form 'the probability that sentence φ has truth-value u is p' as first-order atoms, enabling probabilistic reasoning inside IFOL_B.
Introduced in Section 3 as a new built-in predicate; its truth value is defined through the uncomputable KI, and no independent falsifiable handle or implementation is provided.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Probabilistic Extension of Neuro-Symbolic AGI Robots based on Belnap's Typed Intensional FOL." pith.science (2026). https://pith.science/paper/J5SRVRWP

@misc{pith2026260713073,
  author       = {Pith},
  title        = {Pith review of: Probabilistic Extension of Neuro-Symbolic AGI Robots based on Belnap's Typed Intensional FOL},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/J5SRVRWP}},
  note         = {Machine review of arXiv:2607.13073}
}

abstract

Neuro-symbolic AI based on $IFOL_B$ is a way to combine neural learning and symbolic reasoning to overcome limitations of purely neural systems (like lack of interpretability and logical structure) with formal logical machinery for self-reference. In this paper we expand the cognitive power of $IFOL_B$ by using the probability computation for the currently unknown sentences, based on Nilsson's probability structure for the $IFOL_B$. We introduce the global symmetry transformation that preserves the current knowledge database and logical deduction, and the local one used for real-time decisions about concrete (sub)problems that involve only a very strict subset of $IFOL_B$ predicates. The computation of probability density function $KI$ in both cases, based on the Shannon's maximum information entropy, is provided by neural networks of this probabilistic neuro-symbolic AGI.

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

47 extracted references · 8 linked inside Pith

[1]

: Intensionality and Two-steps Interpretations

Majki´ c,Z. : Intensionality and Two-steps Interpretations. arXiv: 1103.0967, 1–15, (2011)

arXiv 2011
[2]

Walter De Gruyter GmbH, Berlin/Boston, ISBN 978-3-11-099494-0, (20 22)

Majki´ c, Z.: Intensional First Order Logic: from AI to New SQL Big Data. Walter De Gruyter GmbH, Berlin/Boston, ISBN 978-3-11-099494-0, (20 22). 29
[3]

: Intensional FOL: Many-Sorted Extension

Majki´ c,Z. : Intensional FOL: Many-Sorted Extension. ar Xiv:2409.04469 [cs.AI], 1–21, (2024)

arXiv 2024
[4]

: Intensional First Order Logic for Strong-AI Generation of Robots

Majki´ c, Z. : Intensional First Order Logic for Strong-AI Generation of Robots. Journal of Advances in Machine Learning & Artiﬁcial Intelligence, V olume 4, Issue 1, 23–31, (2023)

2023
[5]

: Strong-AI Autoepistemic Robots Build on Int ensional First Order Logic

Majki´ c, Z. : Strong-AI Autoepistemic Robots Build on Int ensional First Order Logic. In F.Zhao, D.Miao(eds), AI-generated Content. AIGC 2023. Communications in Computer and Information Science, vol 1946. Springer, Singapore, 33–58 , (2024)

2023
[6]

: Intensional FOL over Belnap’s Billatice for Strong-AI Robotics

Majki´ c, Z. : Intensional FOL over Belnap’s Billatice for Strong-AI Robotics. arXiv:2508.02774 [cs.LO], 1–29, (2025)

arXiv 2025
[7]

: Universals

Bealer, G. : Universals. The Journal of Philosophy, V ol.9 0, 5–32, (1993)

1993
[8]

: Theories of properties, relations, and propo sitions

Bealer, G. : Theories of properties, relations, and propo sitions. The Journal of Philosophy, V ol.76, 634–648, (1979)

1979

Show all 47 references

[9]

: Intensional logic and epistemic independen cy of intelligent database agents

Majki´ c, Z. : Intensional logic and epistemic independen cy of intelligent database agents. 2nd International Workshop on Philosophy and Informatics ( WSPI 2005), Kaiserslautern, Germany, April 10–13, (2005)

2005
[10]

: Autoepistemic logic programming for reaso ning with inconsistency

Majki´ c, Z. : Autoepistemic logic programming for reaso ning with inconsistency. Interna- tional Symposium on Logic-based Program Synthesis and Tran sformation (LOPSTR), Im- perial College, London,UK, September 7–9, (2005)

2005
[11]

and M.I.Dekhtyar M.I

Dekhtyar A. and M.I.Dekhtyar M.I. : Possible Worlds Sema ntics for Probabilistic Logic Programs. ICLP 2004, 137–148, (2004)

2004
[12]

and Subrahmanian V .S

Udrea O. and Subrahmanian V .S. and Majki´ c Z. : Probabili stic RDF. IEEE Conference on Information Reuse and Integration (IEEE IRI 2006), Septemb er 16–18, Waikoloa, Hawaii, USA, (2006)

2006
[13]

: Temporal Probabilistic Logic Programs: St ate and Revision

Majki´ c, Z. : Temporal Probabilistic Logic Programs: St ate and Revision. International Con- ference on Artiﬁcial Intelligence and Pattern Recognition , AIPR-07, Orlando, Florida, USA, July 9–12, (2007)

2007
[14]

: Probabilistic Deduction and Pattern Recogn ition of Complex Events

Majki´ c Z. : Probabilistic Deduction and Pattern Recogn ition of Complex Events. Interna- tional Conference on Information Security and Privacy (ISP -09), Orlando FL, USA, July 13–16, (2009)

2009
[15]

: Temporal Probabilistic Logic Programs: St ate and Revision

Majki´ c, Z. : Temporal Probabilistic Logic Programs: St ate and Revision. CoRR abs/1103.1061 (2011)

2011 arXiv
[16]

: Probabilistic logic

N.J.Nilsson N.J. : Probabilistic logic. Artif.Intelli gence, 28, 71–88, (1986)

1986
[17]

and Halpern J

Fagin R. and Halpern J. and Megiddo N.: A logic for reasoni ng about probabilities. Informa- tion and Computation, v.87, 78–128, (1990)

1990
[18]

Feller W.: An introduction to Probability Theory and its applications. V ol. I, 2nd ed., Wiley, New Y ork, (1957)

1957
[19]

: Measure Theory

Halmos P . : Measure Theory. V an Nostrand, (1950)

1950
[20]

: Selected works of A.N.Kolmogorov: vol .2 Probability theory and math- ematical statistics

Kolmogorov A.N. : Selected works of A.N.Kolmogorov: vol .2 Probability theory and math- ematical statistics. Moscow: Nauka, by A.N. Shiryayev (Edi tor), (1986)

1986
[21]

and Halpern J.Y .: Uncertainty, Belief, and Prob ability

Fagin R. and Halpern J.Y .: Uncertainty, Belief, and Prob ability. IJCAI 89, 1161–1167, (1989)

1989
[22]

: Intensional Semantics for RDF Data Structur es

Majki´ c Z. : Intensional Semantics for RDF Data Structur es. 12th International Database En- gineering & Applications Systems (IDEAS08), Coimbra, Portugal, 10-13 September, (2008)

2008
[23]

: Constraint Logic Programming and Logic Moda lity for Event’s V alid-time Ap- proximation

Majki´ c Z. : Constraint Logic Programming and Logic Moda lity for Event’s V alid-time Ap- proximation. 2nd Indian International Conference on Artiﬁ cial Intelligence (IICAI-05), De- cember 20-22, Pune, India, (2005)

2005
[25]

: Probabilistic Logic: Many-valuedness and Intensionality

Majki´ c, Z. : Probabilistic Logic: Many-valuedness and Intensionality. CoRR abs/1103.0676 (2011) 30

2011 arXiv
[26]

: Well-Deﬁnedness and Efﬁcient In ference for Probabilistic Logic Programming under the Distribution Semantics

Riguzzi F, and Swift T. : Well-Deﬁnedness and Efﬁcient In ference for Probabilistic Logic Programming under the Distribution Semantics. arXiv:1110 .0631 [cs.AI] (2011)

2011
[27]

and Truszczynski, M

Marek, W. and Truszczynski, M. : Autoepistemic logic. Jo urnal of the ACM 38 (3), 588–618, (1991)

1991
[28]

and Y osinski, J

Nguyen, A. and Y osinski, J. and Clune, J. : Deep neural net works are easily fooled: High conﬁdence predictions for unrecognizable images. In 2015 I EEE Conference on Computer Vision and Pattern Recognition (CVPR), 427–436, (2015)

2015
[29]

: Deep learning: a critical appraisal

Marcus, G. : Deep learning: a critical appraisal. arXiv: 1801.00631 [cs.AI], (2018)

2018 arXiv
[30]

: The purpose put into the machine

Russell, S. : The purpose put into the machine. In Brockma n J (ed) Possible minds: 25 ways of looking at AI, Chap. 3. Penguin Press, New Y ork, 20–32, (20 20)
[31]

and Newell, A

Laird, J.E. and Newell, A. and Rosenbloom, P .S. : SOAR: An architecture for general intel- ligence. In Artiﬁcial Intelligence, V olume 33, Issue 1, 1–64, (1987)

1987
[32]

: Neuro-Symbolic Strong-AI Robots with Closed Knowledge Assumption: Learn- ing and Deductions

Majki´ c, Z. : Neuro-Symbolic Strong-AI Robots with Closed Knowledge Assumption: Learn- ing and Deductions. arXiv:2604.09567 [cs.LO], 1–32, (2026 )

2026 arXiv
[33]

:Business Data Science: Combining Machine Lear ning and Economics to Opti- mize, Automate and Accelerate Business Decisions

Taddy M. :Business Data Science: Combining Machine Lear ning and Economics to Opti- mize, Automate and Accelerate Business Decisions. McGraw- Hill, New Y ork, USA, (2019)

2019
[34]

: Rollout, Policy Iteration, and Distrib uted Reinforcement Learning

Bertsekas S.P . : Rollout, Policy Iteration, and Distrib uted Reinforcement Learning. Athena Scientiﬁc, Belmont, Massachusetts, (2020)

2020
[35]

I srael Journal of Mathematics 2 (1), 1–18

Gaifman H.: Concerning measures in ﬁrst order calculi. I srael Journal of Mathematics 2 (1), 1–18. (1964)

1964
[36]

and Snir M

Gaifman H. and Snir M. : Probabilities over rich language s, testing and randomness. Journal of Symbolic Logic. 47 (3): 495–548. (1982)

1982
[37]

: Random worlds an d maximum entropy

Grove A.J., Halpern J.H., and Koller D. : Random worlds an d maximum entropy. J. Artif. Intell. Res. 2: 33–88. (1994)

1994
[38]

and V encovska A

Paris J.B. and V encovska A. : A note on the inevitability of maximum entropy. Int. J. Approx. Reasoning 14: 183–223. (1990)

1990
[39]

and Kern-Isberner G

Lukasiewicz T. and Kern-Isberner G. : Probabilistic log ic programming under maximum entropy. In Proc. ECSQARU, LNCS 1638, 279–292. (1999)

1999
[40]

: A Mathematical Theory of Communication

Shannon C.E. : A Mathematical Theory of Communication. B ell System Technical Journal 27(3) and 27(4). (1948)

1948
[41]

: A Comparative Study of Neurosymbolic AI Appro aches to Interpretable Logical Reasoning

Chen N.K. : A Comparative Study of Neurosymbolic AI Appro aches to Interpretable Logical Reasoning. arXiv:2508.03366 [cs.AI], (2025)

2025 arXiv
[42]

and Harrison, L

Friston, K., Kilner, J. and Harrison, L. : A free energy pr inciple for the brain. Journal of Physiology-Paris, 100 (1-3): 70–87. (2006)

2006
[43]

: ProbLog: A proba bilistic Prolog and its appli- cation in link discovery

Luc De Raedt, Kimmig A., and Toivonen H. : ProbLog: A proba bilistic Prolog and its appli- cation in link discovery. In IJCAI, pages 2462–2467, (2007)

2007
[44]

and L uc De Raedt : DeepProbLog: Neural Probabilistic Logic Programming

Manhaeve R., Duman´ ci´ c S., Kimmig A., Demeester T. and L uc De Raedt : DeepProbLog: Neural Probabilistic Logic Programming. arXiv:1805.1087 2 [cs.AI], (2018)

2018
[45]

: Sta tistical relational artiﬁcial intelli- gence: Logic, probability, and computation

Luc De Raedt, Kersting K., Natarajan S-, and Poole D. : Sta tistical relational artiﬁcial intelli- gence: Logic, probability, and computation. Synthesis Lectures on Artiﬁcial Intelligence and Machine Learning, 10(2):1–189, (2016). 6 Appendix: From Statistical Physics to Probab...

2016
[46]

Strict Probabilities : - DeepProbLog assigns a probability to a fact (e.g., 0.7::smokes(person) )

Managing Contradictions vs. Strict Probabilities : - DeepProbLog assigns a probability to a fact (e.g., 0.7::smokes(person) ). If sensor data yields ﬂatly conﬂicting conclusions, it dilutes the pr obability distribution. This can lead to logic bottlenecks or failure to compile...
[47]

Extensional Representation : - DeepProbLog relies on classical extensional logic

Intensional vs. Extensional Representation : - DeepProbLog relies on classical extensional logic. It treats predicate s as sets of objects, making it difﬁcult for an agent to separate a real co ncept from its physical instance. - IFOLB applies Intensional Logic. This allows an...
[48]

- IFOLB mitigates this scaling bottleneck by pairing neural networ ks with Local Symmetry Transformations

Execution Bottlenecks and Robotics Scaling : - DeepProbLog relies on algebraic operators that scale poorly during comp lex tasks, generating severe training slowdowns when processi ng highly complex data. - IFOLB mitigates this scaling bottleneck by pairing neural networ ks wi...

Pith tools

Reviewed August 2, 2026 · model on record in the stance chip above.

[1] [1]

: Intensionality and Two-steps Interpretations

Majki´ c,Z. : Intensionality and Two-steps Interpretations. arXiv: 1103.0967, 1–15, (2011)

arXiv 2011

[2] [2]

Walter De Gruyter GmbH, Berlin/Boston, ISBN 978-3-11-099494-0, (20 22)

Majki´ c, Z.: Intensional First Order Logic: from AI to New SQL Big Data. Walter De Gruyter GmbH, Berlin/Boston, ISBN 978-3-11-099494-0, (20 22). 29

[3] [3]

: Intensional FOL: Many-Sorted Extension

Majki´ c,Z. : Intensional FOL: Many-Sorted Extension. ar Xiv:2409.04469 [cs.AI], 1–21, (2024)

arXiv 2024

[4] [4]

: Intensional First Order Logic for Strong-AI Generation of Robots

Majki´ c, Z. : Intensional First Order Logic for Strong-AI Generation of Robots. Journal of Advances in Machine Learning & Artiﬁcial Intelligence, V olume 4, Issue 1, 23–31, (2023)

2023

[5] [5]

: Strong-AI Autoepistemic Robots Build on Int ensional First Order Logic

Majki´ c, Z. : Strong-AI Autoepistemic Robots Build on Int ensional First Order Logic. In F.Zhao, D.Miao(eds), AI-generated Content. AIGC 2023. Communications in Computer and Information Science, vol 1946. Springer, Singapore, 33–58 , (2024)

2023

[6] [6]

: Intensional FOL over Belnap’s Billatice for Strong-AI Robotics

Majki´ c, Z. : Intensional FOL over Belnap’s Billatice for Strong-AI Robotics. arXiv:2508.02774 [cs.LO], 1–29, (2025)

arXiv 2025

[7] [7]

: Universals

Bealer, G. : Universals. The Journal of Philosophy, V ol.9 0, 5–32, (1993)

1993

[8] [8]

: Theories of properties, relations, and propo sitions

Bealer, G. : Theories of properties, relations, and propo sitions. The Journal of Philosophy, V ol.76, 634–648, (1979)

1979